I am currently working on a National Science Foundation research grant with MAT professor and author of Computer Music Tutorial, Curtis Roads and Electrical and Computer Engineer-ing PhD candidate Bob Sturm.

Granular Synthesis is based on the particle-theory of sound proposed initially by Dennis Gabor in the 1950s [ link ]. These grains are parameterized sound units which typically operate on the short-duration time scales (1 - 50ms). Thousands of these grains controlled in statistical cloud distributions or streams can be used to create unique sounds and musical effects.

Until recently, a downside to granular synthesis was the fact that there was no good way to build meaningful granular models of sound. This is not the case with other synthesis techniques. For example, one can get an additive synthesis model for a given sound sample by deriving it's Fourier Transform. The additive synthesis model in this case meaningfully represents the original sound.

New algorithms have emerged in the last few years which build just such a model for granular synthesis. One popular algorithm, called Matching Pursuit, iteratively matches a given signal with grains from a pre-determined dictionary (or set of parameter definitions) of grains. When the algorithm is finished, we are left with a granular model of the original signal which give us totally unique methods of sonic transformations. However, these algorithms are quite slow (much slower than real-time) and there are a number of poorly understood artifacts that need more study. It should be noted that while the analysis is slow, resynthesis is possible in real-time.

In addition to researching novel sound-design and musical applications for this technique, as well as developing software tools to aid in this endeavor, our research is also focused on improving the algorithms. Because of destructive interference between grains in our model, it usually contains more energy than is present in the signal we actually hear. Because we can't hear it, but the energy is there, we call it Dark Energy. Most research in this field is focused on eliminating Dark Energy. We are interested in finding ways we might be able to use it to improve the algorithm or to inform our granular model transformations.

The following are example transformations of a granular synthesis model derived from analyzing a piano excerpt I composed.

The original piano recording:

A straight granular model resynthesis with no transformations:

The following are various transformations of the model:

I am currently working on several other student projects and will post progress information here as they develop or complete. For now, I'll briefly describe them here.

C++ library for procedural music

Many of the techniques which I use to create procedural and interactive music can be implemented within C++. I believe that if we elliminate the need to go to a music-specialized language, and instead use a library that seemlessly exists within middle-ware audio engines, more games will be open to the possibility of procedural music.

Piece for Interactive Music and Dance

I am on a piece which consists of a musical performer and a dancer interacting with a new procedural music system. The dancer, wearing a white costume will interact with the music via a wireless yet simple button. Projected on the dancer will be abstract visualizations which directly represent the state of the music. A musician will control various musical and visual parameters via a Wacom tablet.

VST OSC Plugin

I use Open Sound Control as much as I can and try to avoid MIDI. Unfortunately, that's impossible. MIDI is used by virtually everybody and everything. So I am working on building a few VST plugins that accept and transmit OSC messages and (if you choose), you can map them to MIDI controllers. Needless to say, there are a lot of applications that this would be useful in.